UNSUPERVISED SEGMENTATION OF SMALLHOLDER FIELDS IN MOZAMBIQUE USING PLANETSCOPE IMAGERY

Trees outside forests; in savannas, farmlands, and drylands are largely invisible in global carbon maps. We developed a model that learns to spot trees in satellite photos and estimate how much carbon they store, trained only on rough existing maps without any manual labeling. Our maps turned out far more accurate than the maps it learned from, showing that smarter analysis of satellite imagery can reveal the hidden carbon stored in the world's overlooked trees.

Satellite microwaves help monitor global vegetation, but existing data is too coarse.We enhanced it using higher-resolution information from a biomass map. This produced monthly 5-kilometer resolution data of tropical vegetation optical depth from 2015 to 2021. Our results show the enhanced data accurately captures vegetation dynamics and detects forest disturbances like clearing. By providing this finer-scale dataset, our work supports better local monitoring and protection of ecosystems.

Southwest China has become an important carbon stock due to extensive forest restoration, yet existing biomass datasets are limited in highly heterogeneous landscapes. We developed a fine-grained 30 m aboveground biomass dataset in Southwest China using sub-meter GaoFen imagery and forest structural information. The dataset captures fine-scale spatial heterogeneity, includes trees outside forests, and provides support for regional carbon accounting and restoration efficacy assessment.

This study examines the rate of permafrost degradation, in the form of the transition from intact well-drained palsa to fully thawed and inundated fen at the Stordalen mire, Abisko, Sweden. Across the 14 hectares of the palsa mire, we demonstrate a 5-fold acceleration of the degradation in 2019–2021 compared to previous periods (1970–2014) which might lead to a pool of 12 metric tons of organic carbon exposed annually for the topsoil (23 cm depth), and an increase of ~1.3%/year of GHG emissions.

The Tibetan Plateau, with its diverse vegetation ranging from forests to alpine grasslands, plays a key role in understanding climate change impacts. Existing maps lack detail or miss unique ecosystems. Our research, using advanced satellite technology and machine learning, produced the map TP_LC10-2022. Comparisons with other maps revealed TP_LC10-2022's excellence in capturing local variations. Our map is significant for in-depth ecological studies.